Multi-Modal Wild Salmon Enumeration and Monitoring: A Video and Sonar-based Deployment in Indigenous Rivers of British Columbia

We develop a real-time salmon monitoring system and web application that includes automated counting using computer vision technologies, lowering our partner’s technician resources in performing manual counting. We will deploy the monitoring system in at least nine river locations with small computer hardware capable of deep learning acceleration to run the automated counting system in real-time. The system will have species detection and multi-object tracking to recognize the salmon species and track them across multiple video frames for counting. The sonar system will convert the raw sonar data into video and split them into different regions called stratums before processing. Transferring the sonar data from the rivers while considering the stratums will lower network costs. Low Earth Orbit (LEO) satellite networks will transfer data offsite, where optimization of freezing the congestion window at specific times can help deal with the dynamic satellite movements and improving the throughput of the network.

Faculty Supervisor:

Jiangchuan Liu

Student:

Partner:

Pacific Salmon Foundation

Discipline:

Computer science

Sector:

Agriculture; Other services (except public administration); Professional, scientific and technical services

University:

Simon Fraser University

Program: